In vitro Selection of Allosteric Ribozymes: Theory ... - Famulok, Michael

doi:10.1006/jmbi.2001.4981 available online at http://www.idealibrary.com on J. Mol. Biol. (2001) 312, 885±898In vitro Selection of Allosteric Ribozymes: Theory andExperimental ValidationNicolas Piganeau 1,2 , Vincent Thuillier 2 * and Michael Famulok 1 *1 Kekule Institut fuÈ r OrganischeChemie und Biochemie,Rheinische Friedrich-WilhelmsUniversitaÈt Bonn, Gerhard-Domagk-Strasse 1, 53121Bonn, Germany2 Aventis Pharma, Gencell, 3825Bay Center Place, HaywardCA 94545, USA*Corresponding authorsIn vitro selection techniques offer powerful and versatile methods to isolatenucleic acid sequences with speci®c activities from huge libraries. Wedescribe an in vitro selection strategy for the de novo selection of allostericself-cleaving ribozymes responding to pe¯oxacin and other quinolonederivatives. Within 16 selection cycles, highly sensitive clones respondingto drug levels in the sub-micromolar range were obtained. The morpholinemoiety of the quinolone derivatives was required for inhibition ofthe self-cleavage of the selected ribozymes: modi®cations of the aromaticsystem were tolerated better than modi®cations of the morpholine ring.We also present a theoretical model that analyzes the predicted fractionof ribozymes with a given binding constant and cleavage rate recoveredafter each selection cycle. This model precisely predicts the actual experimentalvalues obtained with the selection procedure. It can thus be usedto determine the optimal conditions for an in vitro selection of an allostericribozyme with a desired dissociation constant and cleavage rate fora given application.# 2001 Academic PressKeywords: ribozymes; aptamers; RNA/drug interaction; allostericselection; selection modelizationIntroductionGene expression may be regulated at every stepof the pathway, from the initiation of transcription,to the activity of the protein itself. The most studiedand seemingly major stage at which geneexpression is regulated in the cell is the initiationof transcription. However, control of geneexpression after this step can occur by regulatingtranscriptional rate, capping, splicing, polyadenylation,cellular localization, stability of the mRNA,translation initiation, translational rate, and ®nallyprotein modi®cation. All of these events can betightly regulated by external signals. 1Controlling the expression of a gene offers abroad range of potential applications, both inresearch and for clinical purposes. 2 First, modulatingthe level of gene expression is an effective wayto study its function. Both knockout and overexpressionof genes of interest have long beenpowerful tools for understanding a gene's biologicalrole. However, in many cases these methodsare limited by several factors: signaling moleculesAbbreviations used: HHR, hammerhead ribozyme.E-mail address of the corresponding authors:m.famulok@uni-bonn.de, vincent.thuillier@aventis.comor pathways are often redundant and compensatefor the under or over-activity of each other, whileconstitutive expression of many genes is cytotoxicor cytostatic. To circumvent such problems, thedevelopment of conditional gene expression systemsthat can be temporally or spatially regulatedis needed. 3We have recently begun to develop systems thatmight allow the regulation of gene expression inresponse to an externally administered regulatorydrug. The basis for such a system is allosteric ribozymes,which can speci®cally cleave their ownRNA in the absence of the regulatory drug and canbe inactived in the presence of the regulatorydrug. 4 Thus, if such ribozymes were suitablyinserted within mRNA transcripts, they wouldlead to its rapid degradation, unless the regulatorydrug is present. Such catalytic RNAs can be rationallydesigned using known ribozymes such as thehammerhead ribozyme (HHR) and regulatorysequences, the so-called aptamers 5,6 that bind smallmolecules by adaptive binding mechanisms. 7,8Alternatively, new effector-binding domains thatactivate or inhibit a ribozyme allosterically can beisolated by in vitro selection methods. 9,10 Using thismethod, Koizumi et al. 9 identi®ed allosteric ribozymesactivated by cyclic nucleotide monophosphatesat concentrations of around 100 mM. Upon0022-2836/01/040885±14 $35.00/0 # 2001 Academic Press

886 Selection of Allosteric Ribozymesactivation, the catalytic rates of these ribozymeswere activated up to 5000-fold, and reached activitiessimilar to the wild-type HHR. 9 Conversely, werecently applied an in vitro selection scheme toobtain allosteric ribozymes that respond to theantibiotic doxycycline. 10 While the level of inhibitionwas around 50-fold, the response could beachieved at very low concentrations with inhibitionconstants as low as 20 nM.In the present study we applied an in vitro selectionscheme similar to that described previously 10to select for an allosteric hammerhead ribozymederivative inhibited by pe¯oxacin, a drug chosen apriori for its potential ability to control geneexpression in vivo. Such a drug must meet severalcriteria: have a good bio-distribution and enter thecells properly, be as innocuous as possible, andpresent features suggesting good RNA bindingproperties. Pe¯oxacin and other quinolone derivativescontain unsaturated rings, as well as ketoneor hydroxyl groups susceptible to interaction withRNA. Pe¯oxacin exhibits relatively low toxicitywith an i.p. LD 50 in rats of 1.5 g/kg, is well distributedwithin the organism, and can permeatecells. 11,12 Quinolones therefore seemed to be goodcandidates for both biological purposes and effectiveinteraction with RNA.The effective control of gene expression by anallosteric ribozyme that is inhibited by a speci®cdrug relies on the self-cleavage rate of the ribozymes(k) and on the af®nity of the drug towardsthe selected ribozyme (K D ). We reasoned that thesetwo critical parameters result from the conditionsapplied during the in vitro selection of allostericribozymes, and therefore that we should be able topredict the ef®ciency of recovery of ribozymeswith given k and K D . We report the description ofsuch a mathematical model, discuss the ®t betweenthe theoretical predictions, and the actual characteristicsof the selected ribozymes and show thatthis model can be used to enhance the effectivenessof the selection of allosteric ribozymes andaptamers.ResultsPool designAllosteric hammerhead ribozymes have beenrationally designed in previous studies by fusingan aptamer RNA onto helix II. 13,14 The stability ofhelix II was previously reported as being crucialfor hammerhead ribozyme activity. 15,16 In addition,this helix is advantageous over positions locateddownstream of the hammerhead ribozymesequence, since the putative binding site of thepotential inhibitor would be very close to the hammerheadcatalytic core (see Figure 1). We thereforechose helix II for positioning the randomizedsequence. Since the desired ribozymes need to beactive in the absence of the drug, it was importantto conserve as much ribozyme activity as possiblein the initial pool. This requirement is in contentionwith the fact that the potential drug-binding pocketshould be as close as possible to the hammerheadcatalytic core for inhibition to occur. In keepingwith these con¯icting requirements, we reducedhelix II to a two base-pair stem and substituted theremainder of stem-loop II by 40 randomizednucleotides (Figure 1).The randomized pool was constructed byemploying two successive large-scale PCRs usingFigure 1. (a) Secondary structure of the transcripts from the initial pool. Helix II was shortened to two base-pairs,and loop II was replaced with a 40 nt random region. The arrow indicates the cleavage site. The chemical linkbetween the biotin moiety and the RNA used during the selection is also shown. (b) Pool ampli®cation scheme. A®rst large-scale PCR was applied to the template Pnp-pool using the ampli®cation primers Pnp-1 and Pnp-rev. In asecond step, the full-length transcriptable pool was produced by PCR with primers Pnp-3 and Pnp-rev. The sequenceof the primers is given in Materials and Methods.

Selection of Allosteric Ribozymes 887the primers shown in Figure 1(b). After synthesisof the double-stranded DNA pool, its complexitywas evaluated as 6 nmol of initial single-strandedoligonucleotides, i.e. 10 15 different molecules. 17This pool was transcriptionally competent, producingRNA molecules of the expected size, andshowed a cleavage ef®ciency of 11 % after overnighttranscription as quanti®ed by denaturing gelelectrophoresis (data not shown). Subsequently,the selection process will actually involve only11 % of the molecules of the initial pool, since ourprocedure selects only ribozymes that are active inthe absence of the drug (see below). No effect of1.0 mM inhibitory molecules on the initial poolcould be detected after transcription under selectionconditions.In vitro selectionIn vitro selection was performed as described. 10Brie¯y, the RNA was transcribed in the presence ofthe inhibitory drug to avoid cleavage, and was biotinylatedat its 5 0 end after T7-transcription. Afterimmobilization on streptavidine agarose, the poolwas incubated in the presence of the inhibitorymolecule. At the end of this incubation, the bounddrug ligand and cleaved ribozyme products wereremoved by denaturing washing steps. Theremaining ribozymes on the column were subjectedto a second incubation, without inhibitor.Finally, the cleaved RNA molecules were eluted,puri®ed, reverse-transcribed, and ampli®ed byPCR before use in the next selection cycle. Thedesign of the PCR primers allowed the restorationof the 5 0 cleaved fraction of those HHR that wereleft bound to the column plus the T7 promoter.The resulting DNA was used for the next selectioncycle.Figure 2 shows the course of the selection indirect comparison to a previously performed selectionthat used doxycycline as the inhibitory drug. 10Remarkably similar results were obtained withpe¯oxacin. The elution pro®le remained constantduring the ®rst three cycles. However, in the fourthcycle a signi®cant decrease in the amount of RNAeluted during the ®rst incubation plus inhibitorwas observed. At the same time, the percentage ofRNA eluted during the second incubationincreased. This result was repeated during cycles 5and 6. However, a control selection cycle in theabsence of inhibitor molecule revealed that theselected pools responded only poorly to theirrespective ligand. For example, in the presence ofdoxycycline, 14 % of the RNA was eluted in selectionstep 4, versus 10 % in the negative controlwithout regulatory drug. To con®rm this result, weassessed the time-course of self-cleavage with thepools after cycle 5 with or without the inhibitormolecules. In the case of the pe¯oxacin pool, aslight inhibition was observed, as illustrated inFigure 3.In contrast, no inhibition at all could be seen inthe pool selected for doxycycline, a phenomenonthat can be explained by the enrichment of parasiticsequences. These are sequences that can foldinto different conformations, some cleavage-competent,some not. After the denaturation-renaturationstep to remove bound ligand and cleavedRNA, some parasitic sequences refold into activecleavage conformations and are eluted togetherwith ligand-inhibited species.Figure 2. Percentages of eluted radioactivity at the end of each selection cycle. The radioactivity eluted after thesecond incubation for the doxycycline (®lled circles) or the pe¯oxacin (®lled squares) selections are shown. Fromcycle 6, a negative control experiment of the respective RNA without regulatory drug in the ®rst incubation step wascarried out to distinguish between enrichment of inhibited and ``parasitic'' ribozymes (open symbols). The tablebelow the graph relates each selection cycle to the respective selection conditions. Arrows indicate selection cyclesampli®ed under mutagenic PCR conditions.

888 Selection of Allosteric RibozymesFigure 3. Cleavage of the pools selected after cycle 5. The cleavage reaction was performed in the absence (®lledcircles) or in the presence (empty circles) of 1.0 mM regulatory drug ((a) doxycycline selection and (b) pe¯oxacinselection). The pool cleavage ef®ciency is severely reduced compared to the wild-type ribozyme (less than 25 %cleaved after one hour). The response to doxycycline is not detectable, and the response to pe¯oxacin is very low.To eliminate parasitic species, we modi®ed theselection procedure from cycle 6 onwards by interruptingthe ®rst incubation step by repetitivewashing steps under denaturation/renaturationconditions. This slight modi®cation of the in vitroselection scheme was based on a theoretical analysis(see Theory). Indeed, after round 7 theenriched pool showed much lower parasiticactivity. After rounds 10 and 13, the complexity ofthe enriched pool was increased by mutagenicPCR. In parallel, the selection stringency was alsoincreased to promote the enrichment of allostericribozymes with improved values of K i and catalyticrate.The self-cleavage activity of the selected poolsfrom cycles 10, 13, and 16 was monitored by timecourseexperiments with different concentrations ofthe inhibitor. Figure 4 directly compares the poolactivitiesfor the previously described doxycycline10 and the now performed pe¯oxacin selections.While the sensitivity of the pooled ribozymes fordoxycycline increased continuously up to cycle 16,where the pool was completely inhibited in thepresence of 1 mM doxycycline, the pe¯oxacin selectionshowed no improvement in the pool responsebetween cycles 13 and 16 but showed inhibitionlevels comparable to those of doxycycline aftercycle 16. Therefore, the majority of the sequencedclones that resulted from the pe¯oxacin-selectionwere taken from cycle 13.Figure 4. Responsiveness of pools from different selection cycles to doxycycline ((a) top row) or pe¯oxacin ((b)lower row). The selected pools were incubated with the indicated amount of the respective drug after cycle 10, 13and 16, and the cleavage activity was monitored. The sensitivity to doxycycline increased gradually during the selection.The sensitivity to pe¯oxacin however did not seem to increase between cycles 13 and 16.

Selection of Allosteric Ribozymes 889Table 1. Single clone analysis after pe¯oxacin selectionClones per class Uninhibited k (min 1 ) Factor of inhibitionClass AP13-28 5 0.34 4.4P16-04 - 0.03 3.5P13-25 2 0.11 0.8P13-06 2 0.15 4.5P13-27 0.15 1.8Class BP13-07 3 0.26 6.1P13-30 2 0.30 5.3Class CP13-03 2 0.26 2.8OrphansP10-01 0.30 1.1P13-15 0.33 13.5P13-19 0.06 4.1P13-24 0.42 3.4Single clone analysis after pe¯oxacin selection. The clones analyzed are shown grouped in classes. The cleavage rate (k) was determinedduring the ®rst three minutes of reaction. The cleavage rate in the presence of 1 mM pe¯oxacin was also determined. Theratio between the uninhibited and inhibited reaction rates is shown (factor of inhibition).Cloning and sequencingThe selected pools were cloned into the vectorpNPG1, and 40 clones were sequenced (Figure 5).Sequence analysis showed that 95 % of thesequences were unique. However, these sequencescould be grouped into different families as shownin Figure 5. The sequences in each group mostprobably originated from one unique sequence inthe initial pool, and the mutagenic PCR gave riseto the intra-group variability. Some groups weresigni®cantly different and were unlikely to haveevolved from the same sequence but neverthelessshowed some common motifs. However, amongthese groups, standard RNA-folding algorithmscould not predict any common secondary structure.18,19Whereas a certain bias towards T could beobserved in the sequenced pool for doxycycline(660 T compared to 443 A, 473 G and 461 C), thisbias was not present in the pe¯oxacin selection(393 T, 367 A, 482 G and 357 C). 10Clone analysis and inhibition of clonesselected by pefloxacinA total of 12 clones were chosen on the basis oftheir representation and were functionally tested.The results obtained are summarized in Table 1.The levels of self-cleavage inhibition observed inthe presence of 1 mM pe¯oxacin were somewhatlower than those observed with the doxycyclineselection, ranging from no inhibition (two clones)to 15-fold inhibition. 10 It is also noteworthy thatthe cleavage rates in the absence of pe¯oxacin aresubstantially lower than the cleavage rate of thewild-type ribozyme from which they are derived(3 min 1 , data not shown)Sensitivity of the clonesFour clones with the highest ratios betweenuninhibited and inhibited rates were chosen for K idetermination. Among them, one yielded irreproduciblekinetics. The other three showed very similarproperties, with K i values ranging from 130 to250 nM (Table 2).These values are about two- to tenfold higherthan those determined for doxycycline. 10 A possibleexplanation is that the clones isolated after thedoxycycline selection were mainly from cycle 16,whereas the clones isolated after the pe¯oxacinselection were from the 13th, and therefore lessstringent, cycle.Specificity of the inhibition and importantfeatures of the pefloxacin moleculeThe two clones showing the lowest K i , P13-15and P13-28, were tested in the presence of 2 mMdifferent derivatives of 1 (Figure 6). These twoclones exhibited the same pattern of recognition forthe different derivatives, as summarized in Figure 6and Table 3. The different modi®cations in theTable 2. Sensitivity of inhibition for three clones after the pe¯oxacin selectionClone Uninhibited k (min 1 ) Pefloxacin K i (nM) Factor of inhibitionP13-07 0.26 0.12 250 170 6P13-15 0.33 0.06 220 100 14P13-28 0.34 0.13 130 70 5K i values for pe¯oxacin, rate constants of the non-inhibited reaction (k), and maximal levels of inhibition are shown.

890 Selection of Allosteric RibozymesFigure 5. Sequences of the clones recovered after the pe¯oxacin selection. The initial random region is shown.Sequences could be grouped into classes, whose consensus sequence is shown in bold. Some classes showed somehomology between each other (class A and B); the conserved regions among class A or B members are underlined.Orphans are sequences that could not be assigned to any class. The clones chosen for further analysis are shown inbold.derivatives had different effects on the inhibition ofthe ribozymesWhen the ¯uorine residue in position 6 was substitutedwith a chlorine atom (9) or the carbonatom in position 8 was substituted with a nitrogenatom (3) no change in inhibition was observed.Similarly, removal of the 4' methyl group (2) or theshortening of ethyl at position 1 to methyl (8) didnot show any effect. However, modi®cations of thepiperazine group, such as addition of an oxygenatom either 3 0 (6) or 4' (4), totally abolished theinhibition effect. Two other derivatives, where thepiperazine group was either substituted for achlorine residue (5) or displaced (7) showed noinhibitory activity.Taken together, these results suggest that thepiperazine ring plays a central role in the inhibitoryfunction of pe¯oxacin 1. However, it is not theonly motif involved in recognition, as displacementof this group on the molecule abolishes activity.The face of the molecule containing the carboxylicacid function (atoms 2 to 5) presents interestingfeatures, which could be involved in the bindingmotif. However, the lack of available derivativesfor this part of the molecule prevented furtherinvestigation.TheoryElimination of parasitic ribozymesThe impact of denaturation/renaturation stepson the recovered fraction (r(x)) of ribozymes insensitiveto pe¯oxacine but folding into both activeand inactive conformations can be computed for agiven ribozyme sequence as:Table 3. Effect of the pe¯oxacin derivatives 1-9 on clones P13-15 and P13-28Entry Compound name Effect Compound1 H 2 O - -2 Pefloxacin ‡‡‡ 13 Norfloxacin ‡‡‡ 24 RP54328 ‡‡‡ 35 RP44992 /‡ 46 RP54240 /‡ 57 RP57606 /‡ 68 RPR105509 - 79 RPR130687 ‡‡‡ 810 RP41983 ‡‡‡ 9Pe¯oxacin (1) and the modi®cation derivatives (2 to 9) refer to the molecules in Figure 6.

Selection of Allosteric Ribozymes 891Figure 6. Structure of pe¯oxacin(1) and its derivatives (2-9).r…x† ˆx…1 x† n…1†where x is the fraction of the molecules foldinginto a self-cleaving conformation and n is the numberof denaturation-renaturation steps:1n ‡ 1The recovered fraction of such ribozymes isexpressed by:is obtained for:r 0 …x† ˆ‰1 …n ‡ 1†xŠ…1 x† n1 ˆ 0…2† 1n ‡ 1rˆ 1 n ‡ 1 1 1 n…4†n ‡ 1x ˆ 1…3†n ‡ 1Therefore the recovered fraction of parasitic ribozymes,r(x), is maximal for the species that fold intoa self-cleaving conformation with a probability of:The maximum fraction of parasitic ribozymes thatcould be eluted after a selection cycle using 12denaturation - renaturation steps is 3 % (equation(4)). This value is an overestimate of the proportionof parasitic ribozymes recovered, because they donot all fold into a self-cleaving conformation with aprobability of 1/(n ‡ 1).

892 Selection of Allosteric RibozymesTherefore, the theoretical analysis supports denaturation-renaturationas an ef®cient means to eliminateparasitic sequences.Outcome of a selection cycleIn order to better understand the outcome of theselection of allosteric ribozymes, we developed atheoretical model, which calculates the enrichmentof allosteric ribozymes with different catalytic andbinding ef®ciencies during a selection cycle.In standard selections (aptamers), the enrichmentin ``winner'' RNA sequences is explained by competitionbetween the RNA species for binding tothe target molecule at the equilibrium. 20 However,in the case of the selection described here, there isno such competition between RNA moleculesbecause the concentration of the regulatory drug isalways kept much higher than the RNA concentration.Rather, the dynamic nature of the enzymaticreaction of self-cleavage drives the selection.To illustrate this, a simple dynamic selection occursif one selects for self-cleaving ribozymes with acleavage rate over 0.69 min 1 (i.e. a half-life of lessthan one minute), assuming a simple exponentialdecrease in the reaction kinetics. If one uses a cleavagetime of one minute, the ribozymes with cleavagerates of exactly 0.69 min 1 will be recoveredat 50 %, because one half of the ribozyme willcleave during one minute and the other half willnot. Ribozymes with cleavage rates of 2 min 1would be recovered at 86 % and ribozymes withcleavage rates of 0.2 min 1 will be recovered at18 %. The inactive ribozymes will be recovered atthe same level as the background. If the backgroundrecovery is 1 %, the enrichment will thereforebe 86-, 50-, 18-, or onefold over thebackground for ribozymes with the respective cleavagerates of 2, 0.69, 0.2, or 0 min 1 . The selectionwill therefore preferentially enrich for ribozymeswith higher cleavage rates.In the case of the present selection, the samekind of reasoning can be applied, with more complexequations. The theoretically recovered fractionis the percentage of the input RNA that will beeluted at the end of the selection cycle for RNAsequences exhibiting a given K D and k (Figure 7(a)).As mentioned earlier, recovered fraction is the keyoutcome of the selection, as it will determine the®nal frequency of one particular sequence after theselection cycle. For example, a sequence with afraction of recovery of 10 % in an overall recoverybackground of 0.5 % will be enriched 20 times.We constructed a model in order to estimate therecovered fraction under the particular conditionsof a selection cycle. Three major hypothesesunderlie this model. First, we suppose that thetranscribed ribozymes undergo self-cleavage followinga ®rst-order kinetics with a constant cleavagerate. Hence, the fraction of uncleavedribozymes decays with time according to a simpleexponential. This assumption applies to step 0 andstep 1 of the selection (Figure 7(a)).The second hypothesis is that only one inhibitormolecule binds each ribozyme, in a reversible manner,and the association/dissociation rates areassumed to be much faster than the cleavage reaction.The ligand-HHR interaction would thereforealways be at equilibrium. This hypothesis relies onthe fact that for most enzyme-ligand associationsthe dissociation rates are higher than 10 s 1 ,whereas the ribozyme cleaving reaction is about5 10 2 s 1 . 16,21 The last major hypothesis in thismodel is that all RNA molecules have the sameability to survive the entire selection process, apartfrom the intended selection, i.e. are transcribed atthe same constant rate, linked on the column andreverse transcribed, with the same ef®ciency. Thishypothesis may not be true during the ®rst roundsof selection. 20Taking all these hypotheses into account, thetheoretical fraction of recovered ribozymes with agiven dissociation constant (K D ) and cleavage rate(k) after one selection cycle was calculated. Formodeling purposes the selection cycle can bereduced to three steps: transcription, incubationwith drug, incubation without drug (Figure 7(a)and (b)). The recovered fraction at the end of theselection cycle is the product of the recovered fractionsat each step.The kinetics of the ribozyme self-cleavage can beexpressed by f that is the fraction of uncleavedribozymes after a time t:f ˆ‰UŠ‰UŠ‡‰CŠ ˆ ektUncleaved RNA ! kCleaved RNA…5†The recovered fraction at the end of the transcription(step 0, Figure 7(a)) is the fraction of theribozyme molecules which remain uncleaved. Witha constant transcription rate K, a transcription timeT 0 , and a drug concentration c 0 , this fraction can becalculated:f 0 ˆ‰UŠ‰UŠ‡‰CŠ ˆ 1 ek appT 0k app T 0Matrix ! K Uncleaved RNA ! kK D mCleaved RNAmUnc: Drug ! k0 Cleav: Drugk app ˆ k K Dc0‡ k 01 ‡ K …7†Dc0If one makes the assumption that k' ˆ 0 (the ribozymecomplexed with the drug is totally inactive),the expression of k app can be simpli®ed:…6†

Selection of Allosteric Ribozymes 893Figure 7. (a) Schematic representationof the selection procedure(see Material and Methods).(b) Model of a selection cycle. Theuncleaved fraction of the ribozymeswith a given k and K D at each stepof the selection is expressed as afunction of the incubation time t.The set of parameters for the actualselection are the duration of theincubations (T 0 , T 1 and T 2 ) and theconcentrations of the regulatorydrug used during the transcriptionand after coupling the ribozymes tothe streptavidin agarose column (c 0and c 1 ). See Theory for details.1k app ˆ k1 ‡ c 0K D…8†Equations (6) and (8) show that inhibited ribozymesbehave like wild-type ribozymes with anapparent cleavage rate (k app ). 16 Another consequenceof this model is that the inhibition constant,K i , and the binding constant, K D, are equal(equation (8)).Similarly, the uncleaved fraction of the ribozymeafter the incubation with the effector (Figure 7,step 1) with an incubation time T 1 and a drug concentrationc 1 , is:f 1 ˆUncleaved RNA ! kK D m‰UŠ‰UŠ‡‰CŠ ˆ ek appT 1Cleaved RNAmUnc: Drug ! k0 Cleav: Drug…9†

894 Selection of Allosteric Ribozymesk app is calculated as in equations (7) and (8) with c 1instead of c 0 .The uncleaved fraction of the ribozyme afterincubation without effector (Figure 7, step 2) withan incubation time T 2 , is:f 2 ˆUncleaved RNA ! k‰UŠ‰UŠ‡‰CŠ ekT 2Cleaved RNA…10†For this step the recovered fraction corresponds tothe cleaved ribozyme fraction that is (1 f 2 ). Consequently,the recovered fraction (r) at the end ofone selection cycle (Figure 7) will be for this particularribozyme species:r ˆ f 0 f 1 …1 f 2 †…11†It is intuitive that for a given cleavage rate, themolecules having the strongest af®nity for theirligand will be selected. Thereby, r is a growingfunction of K D . For a given K D , however, there willbe an intermediate cleavage rate for optimal recovery.Indeed, the ribozyme cleavage rate in the presenceof the ligand (k app ) is proportional to thecleavage rate in the absence of ligand (equation(8)). If k is too high, k app will be so high that themajority of the RNA molecules will cleave duringstep 1 (Figure 7(a)) and thus will be discarded.Conversely, if the cleavage rate is too low, only avery small fraction of the ribozymes will cleaveduring step 2 (Figure 7(a)), and the eluted amountwill be very low. As a consequence, r shows amaximum as a function of k (see below).Fit of the model to experimental dataThe experimental analysis of the individualclones allowed a veri®cation of the pertinence ofthe hypothesis underlying the theoretical model.(i) The kinetics performed on the different clones atvarious concentrations of the regulatory drugshowed a good ®t to the ®rst-order kinetic reaction(simple exponential decay of the uncleaved fractionduring at least two half-lives of the reaction).(ii) for each clone the observed apparent cleavagerate (k app ) in the presence of various concentrationsof the regulatory drug ®tted equation (8) nicely,justifying the hypotheses that each selected ribozymebinds a unique regulatory molecule in areversible fashion and that this association-dissociationis at equilibrium. Furthermore, this alsodemonstrates the hypothesis that k' in equation (7)can be neglected.Under the conditions used during the ®nalrounds of selection, for a determined K D , an optimalk can be calculated (equation (11)). It is strikingthat for the seven clones analyzed from the doxycyclineand the pe¯oxacin selections, the actualcleavage rate is in excellent accordance with thepredicted optimal k (Figure 8) at which maximalrecovery is observed. Therefore, our model seemsto be valid, and explains well the slow self-cleavageobtained for the selected ribozymes.Use of the model for setting the parameters ofthe selectionNext, we show how the model could be used todetermine the optimal conditions for a selectionprotocol, knowing the desired K D and k for a givenapplication. First, we want to ensure that the selectionpressure remains low during the transcriptionreaction and is applied only during the later twoFigure 8. Theoretical recovery after cycles 16 and 13. The curves showing the fractional function of recovery of theribozymes with cleavage rates along the x-axis, for the K i observed for selected clones. The experimentally measuredk values for these clones are indicated by open circles. Except for clone D13-01 (K i ˆ 70 nM) the measured cleavagerates (k values, open circles) very closely correspond to the cleavage rate at which the theoretical fraction of recoveryis maximal. Thus, for these clones the theory almost perfectly re¯ects the experimental results.

Selection of Allosteric Ribozymes 895incubations; therefore we maintained the targetdrug concentration high (1 mM) during the wholetranscription stage. As a consequence, we considerthat no selection is imposed on aptazymes withdissociation constants lower than 1 mM during thetranscription reaction. For these molecules, theselection concentrates on the two later incubationsand the expression of the recovered fraction r canbe simpli®ed with the approximation that f 0 ˆ 1.With this approximation, the catalytic rate withthe maximum level of recovery can be calculated(maximum of r(k)):k opt ˆ 1 ln 1 ‡ c 1 T2‡ 1 …12†T 2 K D T 1For this value of k the fraction of recoverybecomes:2 h i32ln 1 ‡ c 1 T2K D T 1‡ 1r…k opt †ˆexp4 5411 ‡ c 1 T2K D T 111 ‡ c 1K DT2T 1‡ 1…13†The observation of equations (12) and (13) allowsus to draw a strategy for the determination of optimalselection parameters in a selection cycle. r(k opt )is a function of:1 ‡ c 1 T2K D T 1To get a fraction of recovery r(k opt ) ˆ 0.2 for thedesired aptazymes:1 ‡ c 1 T2K D T 1should be approximately 0.7 (numerical resolution).When this parameter is ®xed, k opt is onlydepending on T 2 .If the desired aptazymes are to have catalyticef®ciencies in the range of the wild-type ribozyme(2 min 1 ), the parameter T 2 should be ®xed at16 seconds. Now, with T 2 and:1 ‡ c 1 T2K D T 1®xed, and a desired K D of 50 nM it is possible todetermine T 1 and c 1 . For example one could use aconcentration of drug c 1 ˆ 1 mM and an incubationtime T 1ˆ eight minutes. Another combination givingthe same results could be c 1 ˆ 10 mM and T 1ˆ1.28 hours.It should be noted that the parameters used hereare not optimal for the ®rst round of selection,because a recovered fraction of 20 % would discardmany unique molecules with the appropriate properties.DiscussionWe report the selection of allosteric ribozymesthat are inhibited by pe¯oxacin, an antibiotic of thefamily of ¯uoroquinolones. This class of compoundstarget bacterial DNA gyrase and topoisomeraseIV and stabilize the complexes formedby these enzymes and DNA. 22 Moreover quinoloneshave some af®nity for DNA (K D in the lowmillimolar range) in the presence of magnesiumions. 23 Hammerhead ribozymes are dependent onMg 2‡ for self-cleavage, therefore it is tempting topropose that inhibition of the selected ribozymesby pe¯oxacin occurs through Mg 2‡ displacement.The success of obtaining any biological moleculewith desired characteristics or function depends toa great extent on the initial diversity of the populationand on the power of selection. The advantageof generating the starting material in vitro, aswith functional RNA molecules is that the pool3size and complexity are not limited by biological5 constraints. Pool size and complexity can routinelyreach 10 15 molecules representing orders of magnitudemore than can possibly be obtained in vivo.However, this offers no advantage if suitable selectionparameters cannot be applied, this is why webuilt a mathematical model that describes ourin vitro selection with a restricted set of parameters,which recapitulate the selection parameters. Weshow that once the desired k and K D of the selectedribozymes are known, the model enables one tooptimize the incubation times and the ligand concentrationused at the different steps of the selection.Additionally, with simple calculations wewere able to improve our selection strategy toremove parasitic molecules and obtain better fractionsof molecules with desired characteristics. Theexperimental selection produced doxycycline andpe¯oxacin-sensitive aptazymes with self-cleavagekinetics that correlated remarkably well with theoreticalpredictions.We have demonstrated previously that theselected allosteric ribozymes derived from hammerheadribozyme can be converted into aptamersby a single mutation (cytosine residue 5 0 to thecleavage site into guanosine) that abolishes theself-cleavage activity. 10 Therefore, we believe thatthe selection procedure we report here can beapplied more generally for selecting aptamers, inparticular towards small molecules that shouldotherwise be crosslinked to a solid phase duringthe selection. This gives widespread application tothe predictive model of the selection describedhere.Materials and MethodsPool synthesisThe ®ve primers were synthesized at a micromolarscale on an Expedite-Nucleic Acid Synthesis System(Millipore) following classical phosphoramidite chemistry.Special care was taken that the ampli®cation pri-

896 Selection of Allosteric Ribozymesmers were synthesized and puri®ed before the templateprimer was synthesized, to avoid any contamination byampli®cation primer.The sequences of the primers used were: Pnp-1, 5 0 -CGC GTT GTG TTT ACG CGT CTG ATG-3 0 ; Pnp-2, 5 0 -AGC TGG TAC CTA ATA CGA CTC ACT ATA GGAGCT CGG TAG TGA CGC GTT GTG TTT ACG CGTCTG ATG-3 0 ; Pnp-3, 5 0 -AGC TGG TAC CTA ATA CGACTC ACT ATA GGA GCT CGG TAG TCA CGC GTTGTG TTT ACG CGT CTG ATG-3 0 ; Pnp-rev, 5 0 -ACG TCTCGA GGT AGT TTC GT-3 0 ; Pnp-pool, 5 0 -CGC GTT GTGTTT ACG CGT CTG ATG AGT-N40-ACG AAA CTACCT CGA GAC GT-3 0 .After synthesis primers were cleaved from the resinand deprotected overnight in 1 ml of 32 % ammonia at55 C. Following precipitation with butanol-1, pelletswere puri®ed on denaturing 20 % (Pnp-1 and Pnp-rev)or 12 % (Pnp-2, Pnp-3 and Pnp-pool) PAGE. Bands correspondingto the expected molecular size were excised,crushed and incubated in 14 ml of 0.3 M sodium acetatefor one hour at 65 C and then overnight at room temperature.Eluted nucleotides were separated from gelslurry by ®ltration through glass wool then ethanolprecipitated(sodium acetate). Puri®ed primers wereresuspended in 300 ml of water. Concentration was determinedby measuring absorption at 260 nm.After primer synthesis, large-scale PCR with the primerpool as template plus primers 1 and Rev was performedas follows: total volume 80 ml, template (Pnppool)6.6 nmol, primers (Pnp-1 and Pnp-rev) 80 nmol,MgCl 2 1.5 mM, dNTP 0.2 mM, Gold star reaction buffer1 , DAp Gold star DNA polymerase (Eurogentec) 250units, ®ve cycles (94 C ®ve minutes, 55 C ®ve minutes,72 C seven minutes, in 8 ml aliquots with shaking everytwo minutes). Aliquots (5ml) were taken after each cycleto verify ampli®cation. After PCR, DNA was puri®ed byphenol/chloroform-extraction, ethanol-precipitation andchromatography on a G50 column (0.7 cm 20 cm,using gravity ¯ow, in water) to remove unincorporatednucleotides: 28 nmol (fourfold ampli®cation) of doublestrandedDNA were produced after this PCR as quanti-®ed by absorption at 260 nm and con®rmed on agarosegels.A second step of ampli®cation was done under thesame conditions, using 5 nmol of the ampli®ed product,and primers Pnp-3 and Pnp-rev. The total amount ofrecovered DNA of the appropriate size was 35 nmol(sevenfold ampli®cation).In vitro selectionDuring the selection process, the transcription reactionwas carried out under classical conditions to produceinternally (a- 32 P)-labeled RNA for scintillation countingand quanti®cation of bound and eluted RNA. For selectionpurposes, 1 mM inhibitor molecule and an 8 Mexcess (20 mM ®nal) of guanosine monophosphothioate(GMPS) were used. An 8 M excess of GMPS wasadopted because it gave the best yields of primed molecules.With smaller amounts, the priming ef®ciency waslower while with larger amounts transcription was partiallyinhibited. Quanti®cation of priming was achievedeither through mercury PAGE analysis or by linking tothiopropyl Sepharose. 24 After classical preparationincluding DNase I treatment, precipitation, PAGE puri®cation(8 %) and photometric quanti®cation, the primedRNA was coupled to iodoacetyl-LC-biotin (Pierce). A2 mM solution of primed RNA in 50 mM Tris, 5 mMEDTA (pH 8.3) was incubated with a 200-fold excess ofiodoacetyl-LC-biotin (stock 4 mM in DMF) for 90 minutesat room temperature protected from light withoccasional shaking. This labeling method was usedbecause we observed that direct GMPS linking on thiopropylSepharose was very unstable. 24The reaction products were precipitated (sodium acetate),puri®ed with PAGE and the ®nal amount of RNArecovered was quanti®ed on a photometer. BiotinylatedRNA (1 nmol) was linked for 30 minutes at room temperatureto 250 ml of streptavidin agarose equilibrated incoupling buffer (PBS, 150 mM NaCl, 50 % slurry). Theamount of RNA linked on the column typically rangedbetween 20 and 40 %. To eliminate unlinked species thecolumn was washed thoroughly (six times with alternatively1 ml WA (25 mM Hepes (pH 7.4); 1 M NaCl,5 mM EDTA) and 1 ml WB (3 M urea, 5 mM EDTA))and rinsed with water (®ve times 500 ml). For the ®rstselection incubation, the column material was incubatedin 500 ml selection buffer (40 mM Tris (pH 8.0, 25 C),50 mM NaCl, 2 mM spermidine, 8 mM MgCl 2 ) with theappropriate amount of inhibitor molecule at 37 C withgentle shaking for the appropriate time. The incubationwas initiated upon addition of magnesium. The columnwashing and rinsing sequence was then repeated asbefore and the column was incubated under similar conditionsas before, but without inhibitor molecule. Finally,the cleaved RNA was eluted with WB (two times500 ml). The different fractions (washes and elution) werequanti®ed in a scintillation counter to determine the percentageof eluted RNA. The eluted RNA was puri®edwith two or three phenol/chloroform-extractions, onechloroform extraction and precipitated (sodium acetate)in the presence of glycogen (1 %, v/v). The pellets werewashed two additional times with 70 % ethanol beforeresuspension in 20 ml of water with 200 pmol of Pnp-revprimer. The RNA-oligonucleotide mix was denatured(one minute, 95 C) and mixed with 80 ml of reverse transcriptionmix (5 ml of dNTP 4 mM, 10 ml of MnOAc25 mM, 20 ml of 5x RT-PCR buffer (250 mM bicine/KOH(pH 8.2, 25 C); 575 mM K-acetate; 40 % (v/v) glycerol),43 ml of water, 2 ml of Tth DNA polymerase (two to tenunits)). The total mix was incubated for 30 minutes at72 C (at the same time a control without RNA was performed).The reverse transcription mix was then dilutedinto a 500 ml of PCR reaction under standard conditionswith primers Pnp-3 and Pnp-rev (including a negativecontrol). The number of cycles was calculated using thefollowing equation:n51 ‡ ln 1000 ln…2† …14†xwhere x is the amount of eluted RNA in pmol.The PCR products were analyzed on a 2 % (w/v)agarose gel, before phenol/chloroform-extraction andprecipitation (sodium acetate), 25 % of the resultingDNA was used for the next selection cycle.Mutagenic PCRAt certain points during the selection (after cycles 10and 13), we performed mutagenic PCR using a protocolderived from that described by Cadwell & Joyce. 25 Onthe total DNA molecule, 43 bases could be subject tomutagenic PCR (the other bases are covered by the primers).According to Cadwell & Joyce, 25 this should givean average number of 1.13 mutations per molecule after60 cycles of mutagenic PCR. The number of mutations

Selection of Allosteric Ribozymes 897per individual sequence should follow a Poisson distribution.Hypothetically, the average frequency of unmutagenizedmolecules should be 32 %.The PCR reaction (100 ml) contained 10 mM Tris(pH 8.3), 50 mM KCl, 0.001 % gelatin, 0.2 mM dATPdTTP,1 mM dCTP-dTTP, 7 mM MgCl 2 , 0.5 mM MnCl 2 ,500 pmol of each primer (Pnp-3 and Pnp-rev), 1/50th ofthe DNA ampli®ed in the previous selection cycle, 2 mlof Taq polymerase, for three cycles (94 C 50 seconds,55 C one minute, 72 C one minute) the ®nal temperatureof the block was 55 C and not 4 C as it usually is.The polymerase was added last, when the mix wasalready at 55 C (to avoid enzyme precipitation in thepresence of manganese ions). After the three cycles, 1/8th of the reaction mixture was transferred into a newtube containing the same components (without templateDNA), which was submitted to three new PCR cycles.Again, the polymerase was added just before starting thereaction. After 20 repetitions of this procedure (60 PCRcycles), the mutations were ®xed using a normal PCRprotocol (without manganese ions). This PCR had fourcycles and was done in 800 ml.The amount of DNA was monitored on agarose gelsduring the mutagenic PCR to verify that it did not disappear.A ribozyme activity test was performed at fourdifferent stages during the mutagenic PCR (after 0, 15,30, 45 and 60 PCR cycles). The results showed a 50 %reduction in cleavage ef®ciency after 60 cycles of mutagenicPCR, nicely ®tting with the theoretical 32 % ofunmutagenized molecules (given that some mutationswould be silent). The DNA obtained after 60 cycles ofmutagenic PCR was used for the next selection cycle.KineticsThe DNA templates were transcribed, kinased anddiluted to 1 nM in selection buffer. For the shotgunexperiment, self-cleavage time-course experiments wereperformed without or with 1 mM of the inhibitory drugin selection buffer with 8 mM magnesium at 37 C (selectionconditions). The reaction was initiated uponaddition of magnesium, and aliquots were taken every20 seconds. The aliquots were immediately quenched bymixing with an equal volume of PAGE loading buffer onice. The different aliquots were loaded on a sequencinggel and the uncleaved fraction (radioactivity of theuncleaved band/radioactivity of uncleaved band -‡ radioactivity of the cleaved band) was determined foreach sample after quanti®cation on a PhosphorImager.The cleavage rates (k) were then determined, using anexponential ®tting of the uncleaved ribozyme fractionduring the ®rst two minutes of reaction. For the K i determinations,1 nM kinased ribozyme was incubated withincreasing amounts of the regulatory drug ranging from20 nM to 1 mM, and the reaction rates were calculated asabove. The K i was calculated after ®tting the cleavageratevalues obtained for different drug concentrationswith the following equation (where k is the uninhibitedcleavage rate):kk obs …‰DrugŠ† ˆ1 ‡ ‰DrugŠK i…15†Each experiment was repeated at least two times, andthe values given in the text are the mean and standarddeviation of the observed results.For templates poorly or not inhibited by the drug, thereaction rate was assayed co-transcriptionally asdescribed previously. 16,26AcknowledgementsWe are greatly indebted to Aventis Corporation forproviding compounds 2-9. We thank Michael Blind,Andreas Jenne, Gerhard Sengle and GuÈ nter Mayer forhelpful discussions and suggestions. We thank J. Crouzet,J.F. Mayaux and M. Finer for their support. Thiswork was supported from the Volkswagenstiftung, theFonds der Chemischen Industrie, and Aventis Gencell.References1. Baker, E. J. (1997). mRNA Metabolism and Post-TranscriptionalGene Regulation (Harford, J. B. & Morris,D. R., eds), Wiley-Liss, Inc., New York.2. Gossen, M., Freundlieb, S., Bender, G., Muller, G.,Hillen, W. & Bujard, H. (1995). Transcriptional activationby tetracyclines in mammalian cells. Science,268, 1766-1769.3. Spencer, D. M. (1996). Creating conditionalmutations in mammals. Trends Genet. 12, 181-187.4. Kurz, M. & Breaker, R. R. (1999). In vitro selectionof nucleic acid enzymes. Curr. Top. Microbiol. Immunol.243, 137-158.5. Famulok, M. & Mayer, G. (1999). Aptamers as toolsin molecular biology and immunology. Curr. Top.Microbiol. Immunol. 243, 123-136.6. Famulok, M., Mayer, G. & Blind, M. (2000). Nucleicacid aptamers-from selection in vitro to applicationsin vivo. Accts Chem. Res. 33, 591-599.7. Hermann, T. (2000). Strategies for the design ofdrugs targeting RNA and RNA-protein complexes.Angew. Chem. Int. Ed. 39, 1890-1904.8. Hermann, T. & Patel, D. J. (2000). Adaptive recognitionby nucleic acid aptamers. Science, 287, 820-825.9. Koizumi, M., Soukup, G. A., Kerr, J. N. & Breaker,R. R. (1999). Allosteric selection of ribozymes thatrespond to the second messengers cGMP andcAMP. Nature Struct. Biol. 6, 1062-1071.10. Piganeau, N., Jenne, A., Thuillier, V. & Famulok, M.(2000). An allosteric ribozyme regulated by doxycycline.Angew. Chem. Int. Ed. 39, 4369-4373.11. Bressolle, F., Goncalves, F., Gouby, A. & Galtier, M.(1994). Pe¯oxacin clinical pharmacokinetics. Clin.Pharmacokinet. 27, 418-446.12. Carlier, M. B., Scorneaux, B., Zenebergh, A.,Desnottes, J. F. & Tulkens, P. M. (1990). Cellularuptake, localization and activity of ¯uoroquinolonesin uninfected and infected macrophages.J. Antimicrob. Chemother. 26, (Suppl B), 27-39.13. Araki, M., Okuno, Y., Hara, Y. & Sugiura, Y. (1998).Allosteric regulation of a ribozyme activity throughligand-induced conformational change. Nucl. AcidsRes. 26, 3379-3384.14. Tang, J. & Breaker, R. R. (1997). Rational design ofallosteric ribozymes. Chem. Biol. 4, 453-459.15. Tuschl, T. & Eckstein, F. (1993). Hammerhead ribozymes:importance of stem-loop II for activity. Proc.Natl Acad. Sci. USA, 90, 6991-6994.16. Long, D. M. & Uhlenbeck, O. C. (1994). Kineticcharacterization of intramolecular and intermolecu-

898 Selection of Allosteric Ribozymeslar hammerhead RNAs with stem II deletions. Proc.Natl Acad. Sci. USA, 91, 6977-6981.17. Geiger, A., Burgstaller, P., Von der Eltz, H., Roeder,A. & Famulok, M. (1996). RNA aptamers that bindL-arginine with sub-micromolar dissociation constantsand high enantioselectivity. Nucl. Acids Res.24, 1029-1036.18. Zuker, M., Mathews, D. H. & Turner, D. H. (1999).Algorithms and thermodynamics for RNA secondarystructure prediction: a practical guide. In RNABiochemistry and Biotechnology (Barciszewski, J. &Clark, B. F. C., eds), pp. 11-43, Kluwer AcademicPublishers, Dordrecht.19. Mathews, D. H., Sabina, J., Zuker, M. & Turner,D. H. (1999). Expanded sequence dependence ofthermodynamic parameters improves prediction ofRNA secondary structure. J. Mol. Biol. 288, 911-940.20. Irvine, D., Tuerk, C. & Gold, L. (1991). SELEXION.Systematic evolution of ligands by exponentialenrichment with integrated optimization by non-linearanalysis. J. Mol. Biol. 222, 739-761.21. Fersht, A. (1985). Measurement and magnitude ofenzymatic rate constants. In Enzyme Structure andMechanism, pp. 121-154, W. H. Freeman & Co., NewYork.22. Hooper, D. C. (1999). Mode of action of ¯uoroquinolones.Drugs, 58, (Suppl 2), 6-10.23. Sissi, C., Andreolli, M., Cecchetti, V., Fravolini, A.,Gatto, B. & Palumbo, M. (1998). Mg(2‡)-mediatedbinding of 6-substituted quinolones to DNA: relevanceto biological activity. Bioorg. Med. Chem. 6,1555-1561.24. Sengle, G., Jenne, A., Arora, P. S., Seelig, B.,Nowick, J. S., JaÈschke, A. & Famulok, M. (2000).Synthesis, incorporation ef®ciency, and stability ofdisul®de bridged functional groups at RNA 5 0 -ends.Bioorg. Med. Chem. 8, 1317-1329.25. Cadwell, R. C. & Joyce, G. F. (1994). MutagenicPCR. PCR Methods Appl. 3, S136-140.26. Chowrira, B. M., Pavco, P. A. & McSwiggen, J. A.(1994). In vitro and in vivo comparison of hammerhead,hairpin, and hepatitis delta virus self-processingribozyme cassettes. J. Biol. Chem. 269, 25856-25864.Edited by R. Huber(Received 8 May 2001; received in revised form 24 July 2001; accepted 24 July 2001)